Re: IDR WG Last Call

[Kunihiro Ishiguro <[e-mail]>]

>While I agree that its nice in theory, you have the added difficulty
>of having to maintain additional state for things such as WRD
>and potentially awkward queue mechanisms for outgoing routes.
>As such, you actually implement at least as much, if not more, 
>infrastructure to deal with the additional overhead of having
>a SAFI that covers the unicast+multicast case.
>
>Additionally, policy is now a real pain - and very ambiguous in
>the current spec.  If you say "accept unicast and multicast routes",
>does that mean accept only SAFI 3 or SAFI's 1, 2 and 3?
>
>As Sue has mentioned, treating them as bits works fine for these
>three SAFI's.  But I'd rather not have to.

I agree with this.  

>I'm not going to argue very hard for removing SAFI 3 - we already
>implement it.  But there are complications that don't seem to be
>accounted for with regards to its supposed information overhead reduction.

Yes.  I also implemented SAFI 3.  But I've changed implementation to
not to use it.  I know there was some implementation which support
SAFI 3.  But IMHO some of those implementation already changed their
implementation to not to send SAFI 3.  We've learned it is not good
way to send differenct SAFI route with one NLRI.
-- 
Kunihiro Ishiguro